Model-Driven Architecture of Extreme Learning Machine to Extract Power Flow Features

Area of Science:

• Electrical Engineering
• Computational Science

Background:

• Probabilistic power flow (PPF) is crucial for power system analysis due to increasing uncertainties.
• Existing PPF methods struggle to balance precision and speed, hindering practical applications.

Purpose of the Study:

• To develop a fast and accurate PPF calculation method.
• To address the limitations of current PPF techniques by leveraging machine learning.

Main Methods:

• Designed a model-driven extreme learning machine (ELM) architecture for power flow feature extraction.
• Employed feature decomposition and nonlinearity reduction to simplify learning complexity.
• Optimized hidden node parameters in the ELM for enhanced learning performance.

Main Results:

• The proposed method achieves fast and accurate PPF calculations.
• Simulations on IEEE 57-bus and Polish 2383-bus systems validate the effectiveness.
• The model-driven ELM successfully extracts complex power flow features.

Conclusions:

• The developed ELM-based approach offers a significant improvement for PPF calculations.
• This method enhances the practical applicability of PPF analysis in power systems.
• The integration of physical model properties aids ELM in complex feature learning.